xsk: Introduce AF_XDP buffer allocation API

In order to simplify AF_XDP zero-copy enablement for NIC driver
developers, a new AF_XDP buffer allocation API is added. The
implementation is based on a single core (single producer/consumer)
buffer pool for the AF_XDP UMEM.

A buffer is allocated using the xsk_buff_alloc() function, and
returned using xsk_buff_free(). If a buffer is disassociated with the
pool, e.g. when a buffer is passed to an AF_XDP socket, a buffer is
said to be released. Currently, the release function is only used by
the AF_XDP internals and not visible to the driver.

Drivers using this API should register the XDP memory model with the
new MEM_TYPE_XSK_BUFF_POOL type.

The API is defined in net/xdp_sock_drv.h.

The buffer type is struct xdp_buff, and follows the lifetime of
regular xdp_buffs, i.e.  the lifetime of an xdp_buff is restricted to
a NAPI context. In other words, the API is not replacing xdp_frames.

In addition to introducing the API and implementations, the AF_XDP
core is migrated to use the new APIs.

rfc->v1: Fixed build errors/warnings for m68k and riscv. (kbuild test
         robot)
         Added headroom/chunk size getter. (Maxim/Björn)

v1->v2: Swapped SoBs. (Maxim)

v2->v3: Initialize struct xdp_buff member frame_sz. (Björn)
        Add API to query the DMA address of a frame. (Maxim)
        Do DMA sync for CPU till the end of the frame to handle
        possible growth (frame_sz). (Maxim)

Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200520192103.355233-6-bjorn.topel@gmail.com

include/net/xdp.h

@@ -40,6 +40,7 @@ enum xdp_mem_type {
 	MEM_TYPE_PAGE_ORDER0,     /* Orig XDP full page model */
 	MEM_TYPE_PAGE_POOL,
 	MEM_TYPE_ZERO_COPY,
+	MEM_TYPE_XSK_BUFF_POOL,
 	MEM_TYPE_MAX,
 };
 
@@ -119,7 +120,8 @@ struct xdp_frame *convert_to_xdp_frame(struct xdp_buff *xdp)
 	int metasize;
 	int headroom;
 
-	if (xdp->rxq->mem.type == MEM_TYPE_ZERO_COPY)
+	if (xdp->rxq->mem.type == MEM_TYPE_ZERO_COPY ||
+	    xdp->rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
 		return xdp_convert_zc_to_xdp_frame(xdp);
 
 	/* Assure headroom is available for storing info */

include/net/xdp_sock.h

@@ -31,11 +31,13 @@ struct xdp_umem_fq_reuse {
 struct xdp_umem {
 	struct xsk_queue *fq;
 	struct xsk_queue *cq;
+	struct xsk_buff_pool *pool;
 	struct xdp_umem_page *pages;
 	u64 chunk_mask;
 	u64 size;
 	u32 headroom;
 	u32 chunk_size_nohr;
+	u32 chunk_size;
 	struct user_struct *user;
 	refcount_t users;
 	struct work_struct work;

include/net/xdp_sock_drv.h

@@ -7,6 +7,7 @@
 #define _LINUX_XDP_SOCK_DRV_H
 
 #include <net/xdp_sock.h>
+#include <net/xsk_buff_pool.h>
 
 #ifdef CONFIG_XDP_SOCKETS
 
@@ -101,6 +102,94 @@ static inline u32 xsk_umem_xdp_frame_sz(struct xdp_umem *umem)
 	return umem->chunk_size_nohr;
 }
 
+static inline u32 xsk_umem_get_headroom(struct xdp_umem *umem)
+{
+	return XDP_PACKET_HEADROOM + umem->headroom;
+}
+
+static inline u32 xsk_umem_get_chunk_size(struct xdp_umem *umem)
+{
+	return umem->chunk_size;
+}
+
+static inline u32 xsk_umem_get_rx_frame_size(struct xdp_umem *umem)
+{
+	return xsk_umem_get_chunk_size(umem) - xsk_umem_get_headroom(umem);
+}
+
+static inline void xsk_buff_set_rxq_info(struct xdp_umem *umem,
+					 struct xdp_rxq_info *rxq)
+{
+	xp_set_rxq_info(umem->pool, rxq);
+}
+
+static inline void xsk_buff_dma_unmap(struct xdp_umem *umem,
+				      unsigned long attrs)
+{
+	xp_dma_unmap(umem->pool, attrs);
+}
+
+static inline int xsk_buff_dma_map(struct xdp_umem *umem, struct device *dev,
+				   unsigned long attrs)
+{
+	return xp_dma_map(umem->pool, dev, attrs, umem->pgs, umem->npgs);
+}
+
+static inline dma_addr_t xsk_buff_xdp_get_dma(struct xdp_buff *xdp)
+{
+	struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp);
+
+	return xp_get_dma(xskb);
+}
+
+static inline dma_addr_t xsk_buff_xdp_get_frame_dma(struct xdp_buff *xdp)
+{
+	struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp);
+
+	return xp_get_frame_dma(xskb);
+}
+
+static inline struct xdp_buff *xsk_buff_alloc(struct xdp_umem *umem)
+{
+	return xp_alloc(umem->pool);
+}
+
+static inline bool xsk_buff_can_alloc(struct xdp_umem *umem, u32 count)
+{
+	return xp_can_alloc(umem->pool, count);
+}
+
+static inline void xsk_buff_free(struct xdp_buff *xdp)
+{
+	struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp);
+
+	xp_free(xskb);
+}
+
+static inline dma_addr_t xsk_buff_raw_get_dma(struct xdp_umem *umem, u64 addr)
+{
+	return xp_raw_get_dma(umem->pool, addr);
+}
+
+static inline void *xsk_buff_raw_get_data(struct xdp_umem *umem, u64 addr)
+{
+	return xp_raw_get_data(umem->pool, addr);
+}
+
+static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp)
+{
+	struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp);
+
+	xp_dma_sync_for_cpu(xskb);
+}
+
+static inline void xsk_buff_raw_dma_sync_for_device(struct xdp_umem *umem,
+						    dma_addr_t dma,
+						    size_t size)
+{
+	xp_dma_sync_for_device(umem->pool, dma, size);
+}
+
 #else
 
 static inline bool xsk_umem_has_addrs(struct xdp_umem *umem, u32 cnt)
@@ -212,6 +301,81 @@ static inline u32 xsk_umem_xdp_frame_sz(struct xdp_umem *umem)
 	return 0;
 }
 
+static inline u32 xsk_umem_get_headroom(struct xdp_umem *umem)
+{
+	return 0;
+}
+
+static inline u32 xsk_umem_get_chunk_size(struct xdp_umem *umem)
+{
+	return 0;
+}
+
+static inline u32 xsk_umem_get_rx_frame_size(struct xdp_umem *umem)
+{
+	return 0;
+}
+
+static inline void xsk_buff_set_rxq_info(struct xdp_umem *umem,
+					 struct xdp_rxq_info *rxq)
+{
+}
+
+static inline void xsk_buff_dma_unmap(struct xdp_umem *umem,
+				      unsigned long attrs)
+{
+}
+
+static inline int xsk_buff_dma_map(struct xdp_umem *umem, struct device *dev,
+				   unsigned long attrs)
+{
+	return 0;
+}
+
+static inline dma_addr_t xsk_buff_xdp_get_dma(struct xdp_buff *xdp)
+{
+	return 0;
+}
+
+static inline dma_addr_t xsk_buff_xdp_get_frame_dma(struct xdp_buff *xdp)
+{
+	return 0;
+}
+
+static inline struct xdp_buff *xsk_buff_alloc(struct xdp_umem *umem)
+{
+	return NULL;
+}
+
+static inline bool xsk_buff_can_alloc(struct xdp_umem *umem, u32 count)
+{
+	return false;
+}
+
+static inline void xsk_buff_free(struct xdp_buff *xdp)
+{
+}
+
+static inline dma_addr_t xsk_buff_raw_get_dma(struct xdp_umem *umem, u64 addr)
+{
+	return 0;
+}
+
+static inline void *xsk_buff_raw_get_data(struct xdp_umem *umem, u64 addr)
+{
+	return NULL;
+}
+
+static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp)
+{
+}
+
+static inline void xsk_buff_raw_dma_sync_for_device(struct xdp_umem *umem,
+						    dma_addr_t dma,
+						    size_t size)
+{
+}
+
 #endif /* CONFIG_XDP_SOCKETS */
 
 #endif /* _LINUX_XDP_SOCK_DRV_H */

include/net/xsk_buff_pool.h

@@ -0,0 +1,56 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2020 Intel Corporation. */
+
+#ifndef XSK_BUFF_POOL_H_
+#define XSK_BUFF_POOL_H_
+
+#include <linux/types.h>
+#include <linux/dma-mapping.h>
+#include <net/xdp.h>
+
+struct xsk_buff_pool;
+struct xdp_rxq_info;
+struct xsk_queue;
+struct xdp_desc;
+struct device;
+struct page;
+
+struct xdp_buff_xsk {
+	struct xdp_buff xdp;
+	dma_addr_t dma;
+	dma_addr_t frame_dma;
+	struct xsk_buff_pool *pool;
+	bool unaligned;
+	u64 orig_addr;
+	struct list_head free_list_node;
+};
+
+/* AF_XDP core. */
+struct xsk_buff_pool *xp_create(struct page **pages, u32 nr_pages, u32 chunks,
+				u32 chunk_size, u32 headroom, u64 size,
+				bool unaligned);
+void xp_set_fq(struct xsk_buff_pool *pool, struct xsk_queue *fq);
+void xp_destroy(struct xsk_buff_pool *pool);
+void xp_release(struct xdp_buff_xsk *xskb);
+u64 xp_get_handle(struct xdp_buff_xsk *xskb);
+bool xp_validate_desc(struct xsk_buff_pool *pool, struct xdp_desc *desc);
+
+/* AF_XDP, and XDP core. */
+void xp_free(struct xdp_buff_xsk *xskb);
+
+/* AF_XDP ZC drivers, via xdp_sock_buff.h */
+void xp_set_rxq_info(struct xsk_buff_pool *pool, struct xdp_rxq_info *rxq);
+int xp_dma_map(struct xsk_buff_pool *pool, struct device *dev,
+	       unsigned long attrs, struct page **pages, u32 nr_pages);
+void xp_dma_unmap(struct xsk_buff_pool *pool, unsigned long attrs);
+struct xdp_buff *xp_alloc(struct xsk_buff_pool *pool);
+bool xp_can_alloc(struct xsk_buff_pool *pool, u32 count);
+void *xp_raw_get_data(struct xsk_buff_pool *pool, u64 addr);
+dma_addr_t xp_raw_get_dma(struct xsk_buff_pool *pool, u64 addr);
+dma_addr_t xp_get_dma(struct xdp_buff_xsk *xskb);
+dma_addr_t xp_get_frame_dma(struct xdp_buff_xsk *xskb);
+void xp_dma_sync_for_cpu(struct xdp_buff_xsk *xskb);
+void xp_dma_sync_for_device(struct xsk_buff_pool *pool, dma_addr_t dma,
+			    size_t size);
+
+#endif /* XSK_BUFF_POOL_H_ */

include/trace/events/xdp.h

@@ -287,7 +287,8 @@ TRACE_EVENT(xdp_devmap_xmit,
 	FN(PAGE_SHARED)		\
 	FN(PAGE_ORDER0)		\
 	FN(PAGE_POOL)		\
-	FN(ZERO_COPY)
+	FN(ZERO_COPY)		\
+	FN(XSK_BUFF_POOL)
 
 #define __MEM_TYPE_TP_FN(x)	\
 	TRACE_DEFINE_ENUM(MEM_TYPE_##x);

net/core/xdp.c

@@ -17,6 +17,7 @@
 #include <net/xdp.h>
 #include <net/xdp_priv.h> /* struct xdp_mem_allocator */
 #include <trace/events/xdp.h>
+#include <net/xdp_sock_drv.h>
 
 #define REG_STATE_NEW		0x0
 #define REG_STATE_REGISTERED	0x1
@@ -361,7 +362,7 @@ EXPORT_SYMBOL_GPL(xdp_rxq_info_reg_mem_model);
  * of xdp_frames/pages in those cases.
  */
 static void __xdp_return(void *data, struct xdp_mem_info *mem, bool napi_direct,
-			 unsigned long handle)
+			 unsigned long handle, struct xdp_buff *xdp)
 {
 	struct xdp_mem_allocator *xa;
 	struct page *page;
@@ -390,6 +391,11 @@ static void __xdp_return(void *data, struct xdp_mem_info *mem, bool napi_direct,
 		xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params);
 		xa->zc_alloc->free(xa->zc_alloc, handle);
 		rcu_read_unlock();
+		break;
+	case MEM_TYPE_XSK_BUFF_POOL:
+		/* NB! Only valid from an xdp_buff! */
+		xsk_buff_free(xdp);
+		break;
 	default:
 		/* Not possible, checked in xdp_rxq_info_reg_mem_model() */
 		break;
@@ -398,19 +404,19 @@ static void __xdp_return(void *data, struct xdp_mem_info *mem, bool napi_direct,
 
 void xdp_return_frame(struct xdp_frame *xdpf)
 {
-	__xdp_return(xdpf->data, &xdpf->mem, false, 0);
+	__xdp_return(xdpf->data, &xdpf->mem, false, 0, NULL);
 }
 EXPORT_SYMBOL_GPL(xdp_return_frame);
 
 void xdp_return_frame_rx_napi(struct xdp_frame *xdpf)
 {
-	__xdp_return(xdpf->data, &xdpf->mem, true, 0);
+	__xdp_return(xdpf->data, &xdpf->mem, true, 0, NULL);
 }
 EXPORT_SYMBOL_GPL(xdp_return_frame_rx_napi);
 
 void xdp_return_buff(struct xdp_buff *xdp)
 {
-	__xdp_return(xdp->data, &xdp->rxq->mem, true, xdp->handle);
+	__xdp_return(xdp->data, &xdp->rxq->mem, true, xdp->handle, xdp);
 }
 EXPORT_SYMBOL_GPL(xdp_return_buff);
 

net/xdp/Makefile

@@ -1,3 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0-only
 obj-$(CONFIG_XDP_SOCKETS) += xsk.o xdp_umem.o xsk_queue.o xskmap.o
+obj-$(CONFIG_XDP_SOCKETS) += xsk_buff_pool.o
 obj-$(CONFIG_XDP_SOCKETS_DIAG) += xsk_diag.o

net/xdp/xdp_umem.c

@@ -245,7 +245,7 @@ static void xdp_umem_release(struct xdp_umem *umem)
 	}
 
 	xsk_reuseq_destroy(umem);
-
+	xp_destroy(umem->pool);
 	xdp_umem_unmap_pages(umem);
 	xdp_umem_unpin_pages(umem);
 
@@ -390,6 +390,7 @@ static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
 	umem->size = size;
 	umem->headroom = headroom;
 	umem->chunk_size_nohr = chunk_size - headroom;
+	umem->chunk_size = chunk_size;
 	umem->npgs = size / PAGE_SIZE;
 	umem->pgs = NULL;
 	umem->user = NULL;
@@ -415,11 +416,21 @@ static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
 	}
 
 	err = xdp_umem_map_pages(umem);
-	if (!err)
-		return 0;
+	if (err)
+		goto out_pages;
 
-	kvfree(umem->pages);
+	umem->pool = xp_create(umem->pgs, umem->npgs, chunks, chunk_size,
+			       headroom, size, unaligned_chunks);
+	if (!umem->pool) {
+		err = -ENOMEM;
+		goto out_unmap;
+	}
+	return 0;
 
+out_unmap:
+	xdp_umem_unmap_pages(umem);
+out_pages:
+	kvfree(umem->pages);
 out_pin:
 	xdp_umem_unpin_pages(umem);
 out_account: